Improve semantic information for types compiled by dotty

[bjaglin]

The SymbolTable required by core for finding information that is not available in SemanticDB is injected by cli using a helper from reflect, which is currently backed by scalameta's internal semanticdb-scalac-core which was written for Scala 2 class files.

scalafix/scalafix-core/src/main/scala/scalafix/internal/v1/InternalSemanticDoc.scala

Lines 68 to 82 in c6d7411

	def info(sym: Symbol): Option[SymbolInformation] = {
	if (sym.isNone) {
	None
	} else if (sym.isLocal) {
	locals.get(sym.value).map(new SymbolInformation(_)(this))
	} else {
	val fromGlobalSymtab =
	symtab.info(sym.value).map(new SymbolInformation(_)(this))
	def fromTextDocument(): Option[SymbolInformation] =
	textDocument.symbols
	.find(_.symbol == sym.value)
	.map(new SymbolInformation(_)(this))
	fromTextDocument().orElse(fromGlobalSymtab)
	}
	}

scalafix/scalafix-cli/src/main/scala/scalafix/internal/v1/Args.scala

Lines 200 to 212 in c6d7411

	def configuredSymtab: Configured[SymbolTable] = {
	Try(
	ClasspathOps.newSymbolTable(
	classpath = validatedClasspath,
	out = out
	)
	) match {
	case Success(symtab) =>
	Configured.ok(symtab)
	case Failure(e) =>
	ConfError.message(s"Unable to load symbol table: ${e.getMessage}").notOk
	}
	}

scalafix/scalafix-reflect/src/main/scala/scalafix/internal/reflect/ClasspathOps.scala

Lines 27 to 33 in c6d7411

	def newSymbolTable(
	classpath: Classpath,
	parallel: Boolean = false,
	out: PrintStream = System.out
	): SymbolTable = {
	GlobalSymbolTable(classpath, includeJdk = true)
	}

We need to migrate to https://github.com/scalacenter/tasty-query or equivalent to provide exhaustive information about types compiled by dotty. While doing that, we should drop the need to using Scalameta internal structures as a contract between core & cli.

[tgodzik]

An alternative is to use info method that we added to the presentation compiler. It provides a similar interface as semanticdb.

[bjaglin]

Interesting! Do you know what's the PC overhead? In terms of start up time (I guess in Metals it's warm no it does not matter much) and RAM footprint?

[tgodzik]

The overhead shouldn't be too big, but it's hard to say. It's basically the addition of the compiler, which would already be added when I manage to finish the inferred type PR.